# Airless Node — Offline LLM Evaluation (Public Portfolio Demo)

A **sanitized, standalone example** illustrating the defensive evaluation component of a local-first Red Team / Blue Team LLM testing concept.

## What it demonstrates

- Python-based offline response checks
- Structured results suitable for logging and dashboard integration
- Batch processing and unit tests
- Dependency-free execution; no credentials, private endpoints, network calls, or real user data

## Run

```bash
python src/demo.py
python -m unittest discover -s tests -v
```

## Architecture (conceptual)

```text
Red-team test cases (synthetic) -> Local model under test
                                        |
                                        v
                               Response evaluator
                                        |
                                        v
                             Structured findings/logs
                                        |
                                        v
                             Blue-team review/dashboard
```

## Scope and provenance

This is **portfolio-safe illustrative code**, not the original private application or a claim of deployed multi-node functionality. It was prepared from a review of supplied project snippets, which included TypeScript helpers for zero-width text encoding, timing, and an API evaluator; a UI component export index; and Python package wiring. Those private files are intentionally **not** reproduced here. This demo does not implement model inference, network dispatch, distributed scheduling, or comprehensive prompt-injection detection.

## Privacy and responsible use

Use synthetic test strings and models you own or are authorized to test. Keep raw prompt data, session logs, API credentials, device identifiers, and internal infrastructure out of public repositories.

## License

MIT (see LICENSE).